# Formulating a function on Matlab for the Shannon interpolation formula

I am trying to formulate an algorithm for applying to Shannon interpolation formula to the discrete signal

$$x[n]=\frac{c^2}{4}\int\limits_{0}^{n T}y\left(\tfrac{c}{2}s\right)\,ds,$$

where $c$ is constant. Now, if $\frac{1}{T}=f_{s}>2B$, where $B$ is the band-limit of $X(f)$, then we can reconstruct the continuous time signal as:

$$x_{\text{const}}(t)=\sum\limits_{n=-\infty}^{\infty}\frac{c^2}{4}{\int\limits_{0}^{n T}y\left(\tfrac{c}{2}s\right)\,ds}\cdot\text{sinc}\left(\frac{t-n T}{T}\right)$$

function x=interpolate(y,N,T,c)
x=0;
for n=1:N
x=x+((c^2)/2)*integral(y,0,n.*T).*sinc((t-n.*T)./T);
end


However, I am having trouble with incorporating the t variable here. Setting it as global and running interpolate(y,20,20,1) also produces nothing to note.

Note that the continuous time signal in question is given by

$$x_{\text{true}}(t)=\frac{c^2}{4}\int_{0}^{t}y\left(\tfrac{c}{2}s\right)\,ds.$$

• Can you give a reference, where you have these equations from? To me, Shannon interpolation is simply sampling the original signal without an integral. Further, I believe your integral should run from $\int_{(n-1)T}^{nT}ds$ and I have no idea, what the variable $c$ should define (I would always assume $c=2$ looking at these equations). – Maximilian Matthé Feb 7 '17 at 18:53
• @MaximilianMatthé See this article here: en.wikipedia.org/wiki/… Also, $c$ is a constant, although I thought it'd be interesting to experiment whether I could vary it. However, yes, I also went with $c=2$ for graph plotting, etc. – Jason Born Feb 7 '17 at 18:55
• Well, actually in this Wikipedia article, I cant find the integral you are stating there. – Maximilian Matthé Feb 7 '17 at 19:20
• @MaximilianMatthé I apologise, the integral is due to the fact that the continuous time signal is given by $$x(t):=\frac{c^2}{4}\int_{0}^{t}y\left(\frac{cs}{2}\right)\,ds,$$ which we can sample to get $x[n]$. – Jason Born Feb 7 '17 at 20:11

Given your samples $x[n]$, here's a code to perform the interpolation by the sinc-functions:

F = 1000; % the sampling frequency used for drawing the continuous function
Fs = 50; % the sampling frequency
T = 1/Fs; % the sampling time

t = 0:1/F:1; % the time-samples for "continuous" plotting
ts = 0:T:1;  % the time at the sample points

% the original function. In your case, this would be integral expression
x = @(t) sin(2*pi*10*t) + cos(2*pi*11*t);
xn = x(ts);  % the discrete sequence

% perform interpolation from xn to x:
interpolated = 0;
for n=1:length(ts)
% n-1 due to the 1-based indexing of matlab
interpolated = interpolated + xn(n) * sinc((t-(n-1)*T)/T);
end

hold off;
plot(t, x(t), 'r', 'linewidth', 3); hold on;  % plot continuous
stem(ts, xn);  hold on; % plot sampled
plot(t, interpolated, 'k-o'); % plot interpolated on top of red curve As a bonus, here's what happens when the signal is not enough bandlimited: I changed the frequency of the first sine from $sin(2\pi 10 t)$ to $sin(2\pi 30 t)$: Note: The Shannon-Interpolation requires bandlimited signals, so in particular it does not work exactly with time-limeted signals (since these are never band-limited). So, you will see some deviations especially at the edge of the signal.

• it's a lotta work for a sorta parochial problem (i dunno what the integral was doing in there either). – robert bristow-johnson Feb 8 '17 at 6:38